def __init__(self,
p_value,
test_statistic,
name=None,
test_name=None,
**kwargs):
self.p_value = p_value
self.test_statistic = test_statistic
self.test_name = test_name
self._p_value = utils._to_1d_array(p_value)
self._test_statistic = utils._to_1d_array(test_statistic)
assert len(self._p_value) == len(self._test_statistic)
if name is not None:
self.name = utils._to_list(name)
assert len(self.name) == len(self._test_statistic)
else:
self.name = None
for kw, value in kwargs.items():
setattr(self, kw, value)
kwargs["test_name"] = test_name
self._kwargs = kwargs
def survival_function_at_times(self, times, label=None) -> pd.Series:
"""
Return a Pandas series of the predicted survival value at specific times
Parameters
-----------
times: iterable or float
label: str
"""
label = coalesce(label, self._label)
return pd.Series(self.predict(times), index=_to_1d_array(times), name=label)
def cumulative_density_at_times(self, times, label=None) -> pd.Series:
"""
Return a Pandas series of the predicted cumulative density at specific times
Parameters
-----------
times: iterable or float
Returns
--------
pd.Series
"""
label = coalesce(label, self._label)
return pd.Series(1 - self.predict(times), index=_to_1d_array(times), name=label)
def proportional_hazard_test(fitted_cox_model,
training_df,
time_transform="rank",
precomputed_residuals=None,
**kwargs) -> StatisticalResult:
"""
Test whether any variable in a Cox model breaks the proportional hazard assumption. This method uses an approximation
that R's ``survival`` use to use, but changed it in late 2019, hence there will be differences here between lifelines and R.
Parameters
----------
fitted_cox_model: CoxPHFitter
the fitted Cox model, fitted with `training_df`, you wish to test. Currently only the CoxPHFitter is supported,
but later CoxTimeVaryingFitter, too.
training_df: DataFrame
the DataFrame used in the call to the Cox model's ``fit``.
time_transform: vectorized function, list, or string, optional (default='rank')
{'all', 'km', 'rank', 'identity', 'log'}
One of the strings above, a list of strings, or a function to transform the time (must accept (time, durations, weights) however). 'all' will present all the transforms.
precomputed_residuals: DataFrame, optional
specify the scaled Schoenfeld residuals, if already computed.
kwargs:
additional parameters to add to the StatisticalResult
Notes
------
R uses the default `km`, we use `rank`, as this performs well versus other transforms. See
http://eprints.lse.ac.uk/84988/1/06_ParkHendry2015-ReassessingSchoenfeldTests_Final.pdf
References
-----------
- http://eprints.lse.ac.uk/84988/1/06_ParkHendry2015-ReassessingSchoenfeldTests_Final.pdf
- "Extending the Cox Model"
- https://github.com/therneau/survival/commit/5da455de4f16fbed7f867b1fc5b15f2157a132cd#diff-c784cc3eeb38f0a6227988a30f9c0730R36
"""
if "transform" in kwargs:
warnings.warn(
"Found 'transform' keyword being set. Did you mean to set 'time_transform' instead?",
UserWarning)
events, durations, weights = fitted_cox_model.event_observed, fitted_cox_model.durations, fitted_cox_model.weights
n_deaths = events.sum()
if precomputed_residuals is None:
scaled_resids = fitted_cox_model.compute_residuals(
training_df, kind="scaled_schoenfeld")
else:
scaled_resids = precomputed_residuals
scaled_resids = (
fitted_cox_model.compute_residuals(training_df, kind="schoenfeld").dot(
fitted_cox_model.variance_matrix_) * n_deaths)
def compute_statistic(times, resids, n_deaths):
demeaned_times = times - times.mean()
T = (demeaned_times.values[:, None] * resids.values).sum(0)**2 / (
n_deaths * (fitted_cox_model.standard_errors_**2) *
(demeaned_times**2).sum())
return T
if time_transform == "all":
time_transform = list(TimeTransformers.TIME_TRANSFOMERS.keys())
if isinstance(time_transform, list):
result = StatisticalResult([], [], [])
# yuck
for transform_name, transform in ((_, TimeTransformers().get(_))
for _ in time_transform):
times = transform(durations, events, weights)[events.values]
T = compute_statistic(times, scaled_resids, n_deaths)
p_values = utils._to_1d_array(
[_chisq_test_p_value(t, 1) for t in T])
result += StatisticalResult(
p_values,
T,
name=[(c, transform_name)
for c in fitted_cox_model.params_.index],
test_name="proportional_hazard_test",
null_distribution="chi squared",
degrees_of_freedom=1,
model=str(fitted_cox_model),
**kwargs)
else:
time_transformer = TimeTransformers().get(time_transform)
assert callable(
time_transformer
), "time_transform must be a callable function, or a string: {'rank', 'km', 'identity', 'log'}."
times = time_transformer(durations, events, weights)[events.values]
T = compute_statistic(times, scaled_resids, n_deaths)
p_values = utils._to_1d_array([_chisq_test_p_value(t, 1) for t in T])
result = StatisticalResult(
p_values,
T,
name=fitted_cox_model.params_.index.tolist(),
test_name="proportional_hazard_test",
time_transform=time_transform,
null_distribution="chi squared",
degrees_of_freedom=1,
model=str(fitted_cox_model),
**kwargs)
return result
def proportional_hazard_test(fitted_cox_model,
training_df,
time_transform="rank",
precomputed_residuals=None,
**kwargs):
"""
Test whether any variable in a Cox model breaks the proportional hazard assumption.
Parameters
----------
fitted_cox_model: CoxPHFitter
the fitted Cox model, fitted with `training_df`, you wish to test. Currently only the CoxPHFitter is supported,
but later CoxTimeVaryingFitter, too.
training_df: DataFrame
the DataFrame used in the call to the Cox model's ``fit``.
time_transform: vectorized function, list, or string, optional (default='rank')
{'all', 'km', 'rank', 'identity', 'log'}
One of the strings above, a list of strings, or a function to transform the time (must accept (time, durations, weights) however). 'all' will present all the transforms.
precomputed_residuals: DataFrame, optional
specify the scaled schoenfeld residuals, if already computed.
kwargs:
additional parameters to add to the StatisticalResult
Returns
-------
StatisticalResult
Notes
------
R uses the default `km`, we use `rank`, as this performs well versus other transforms. See
http://eprints.lse.ac.uk/84988/1/06_ParkHendry2015-ReassessingSchoenfeldTests_Final.pdf
"""
events, durations, weights = fitted_cox_model.event_observed, fitted_cox_model.durations, fitted_cox_model.weights
deaths = events.sum()
if precomputed_residuals is None:
scaled_resids = fitted_cox_model.compute_residuals(
training_df, kind="scaled_schoenfeld")
else:
scaled_resids = precomputed_residuals
def compute_statistic(times, resids):
times -= times.mean()
T = (times.values[:, None] * resids.values).sum(0)**2 / (
deaths * np.diag(fitted_cox_model.variance_matrix_) *
(times**2).sum())
return T
if time_transform == "all":
time_transform = list(TimeTransformers.TIME_TRANSFOMERS.keys())
if isinstance(time_transform, list):
result = StatisticalResult([], [], [])
# yuck
for transform_name, transform in ((_, TimeTransformers().get(_))
for _ in time_transform):
times = transform(durations, events, weights)[events.values]
T = compute_statistic(times, scaled_resids)
p_values = _to_1d_array([chisq_test(t, 1) for t in T])
result += StatisticalResult(
p_values,
T,
name=[(c, transform_name)
for c in fitted_cox_model.params_.index],
test_name="proportional_hazard_test",
null_distribution="chi squared",
degrees_of_freedom=1,
**kwargs)
else:
time_transformer = TimeTransformers().get(time_transform)
assert callable(
time_transformer
), "time_transform must be a callable function, or a string: {'rank', 'km', 'identity', 'log'}."
times = time_transformer(durations, events, weights)[events.values]
T = compute_statistic(times, scaled_resids)
p_values = _to_1d_array([chisq_test(t, 1) for t in T])
result = StatisticalResult(
p_values,
T,
name=fitted_cox_model.params_.index.tolist(),
test_name="proportional_hazard_test",
time_transform=time_transform,
null_distribution="chi squared",
degrees_of_freedom=1,
**kwargs)
return result
